Article type
Year
Abstract
Background: Cochrane recommends supplementing journal articles with other data sources to obtain information about trials, however, no clear guidance is given when multiple sources provide contradictory information.
Objective: To compare PICO (participant, intervention, comparator, and outcome) information and risk of bias (ROB) assessment from multiple sources for two case examples: gabapentin for neuropathic pain and quetiapine for bipolar depression.
Methods: We identified eligible reports by searching bibliographic databases, trial registers, conference proceedings, FDA reviews, and reference lists. We also used unpublished documents available from litigation (i.e. internal company documents called Clinical Study Reports or CSRs). Two independent reviewers completed each of the following tasks, and handled disagreements by discussion: screening, data extraction, and ROB assessment. For both cases, we compared the following items across sources: condition, number of participants randomized, sex and age of participants, interventions, comparators, length of follow-up, conflicts of interest, and items from the Cochrane ROB tool.
Results: We identified 21 gabapentin trials and seven quetiapine trials. Most trials were presented in multiple reports. Among multiple reports of the same trial (14 gabapentin trials; six quetiapine trials), we identified substantive discrepancies in the number of groups and participants randomized across reports, and in the descriptions of administration methods and doses of interventions and comparators. For all other aspects of PICO and ROB we examined, we identified differences in completeness of information, but not contradictory information. CSRs, typically thousands of pages in length, can yield more complete information, but require considerable resources for review, double data abstraction, and reconciliation.
Conclusions: In these two case examples, we found that different sources can yield substantively different information and varying levels of completeness on PICO and ROB assessment for trials. Although CSRs can yield additional information about PICO and ROB, abstracting data from them is time-intensive.
Objective: To compare PICO (participant, intervention, comparator, and outcome) information and risk of bias (ROB) assessment from multiple sources for two case examples: gabapentin for neuropathic pain and quetiapine for bipolar depression.
Methods: We identified eligible reports by searching bibliographic databases, trial registers, conference proceedings, FDA reviews, and reference lists. We also used unpublished documents available from litigation (i.e. internal company documents called Clinical Study Reports or CSRs). Two independent reviewers completed each of the following tasks, and handled disagreements by discussion: screening, data extraction, and ROB assessment. For both cases, we compared the following items across sources: condition, number of participants randomized, sex and age of participants, interventions, comparators, length of follow-up, conflicts of interest, and items from the Cochrane ROB tool.
Results: We identified 21 gabapentin trials and seven quetiapine trials. Most trials were presented in multiple reports. Among multiple reports of the same trial (14 gabapentin trials; six quetiapine trials), we identified substantive discrepancies in the number of groups and participants randomized across reports, and in the descriptions of administration methods and doses of interventions and comparators. For all other aspects of PICO and ROB we examined, we identified differences in completeness of information, but not contradictory information. CSRs, typically thousands of pages in length, can yield more complete information, but require considerable resources for review, double data abstraction, and reconciliation.
Conclusions: In these two case examples, we found that different sources can yield substantively different information and varying levels of completeness on PICO and ROB assessment for trials. Although CSRs can yield additional information about PICO and ROB, abstracting data from them is time-intensive.